Writing an AI Playbook for Oncology Practices: Larry Bilbrey

Writing guidance on artificial intelligence (AI) that is comprehensive without steering practices toward specific products was among the hardest parts of developing a playbook for community practices, and technology gaps, not workflows, remain the biggest challenge in adding AI to oncology care, according to Larry Bilbrey of OneOncology.

In an interview with The American Journal of Managed Care^®, Bilbrey discussed the challenges of developing the playbook, the importance of writing it in physician language, and the technology gaps that persist as AI enters oncology workflows.

Why the Playbook Avoids Naming Products

Bilbrey described a committee effort to draw the line between general AI recommendations and very prescriptive ones. That line was hard to hold, because practices regularly ask for advice on specific solutions, he explained. An AI scribe is a common example, and practices want to know which one is best.

"Every practice is different. Every physician is different," Bilbrey said. Because of that, the committee aimed to make the playbook as comprehensive as possible without recommending a product on the grounds that it is the best in the market. Instead, the playbook tells practices what to consider when making decisions. That includes what to think about when writing contracts and what to think about 3 to 5 years out.

A second challenge was making sure the playbook made sense to other physicians. Bilbrey said the committee included many physicians, which helped put the material in their language. That matters because community practices are mostly owned and run by physicians, so they have to understand what the playbook says before they can act on it.

Where Technology Gaps Remain

Bilbrey said technology, not workflow, is the greater obstacle. Pathology and CT reports often arrive by fax from outside imaging centers, and a physician must then find the document and read it. According to Bilbrey, early AI has not been able to pull in those documents and do anything useful with them.

That is starting to change. He said the organization now has AI built to find those pathology and imaging reports, and the process is improving with each iteration. "We're getting there. We're getting better, and we're iterating on that process," Bilbrey said.

Workflows, by comparison, are well established. Bilbrey noted that the organization has been doing this work for 50 years and that things move smoothly 98% of the time, even with the occasional hiccup. Adding AI changes that picture.

"Now that we're adding AI to the mix, these technology gaps are very real," he said.